Journal of Economic and Social Measurement 20 (1994) 79-94 79IOS Press

COMPARING PUBLIC AND PRIVATEEARNINGS USING STATE WAGE SURVEYS*

Dale BELMAN, Thomas E. FRANKLIN and John S. HEYWOODDepartment of Economics, University of Wisconsin,

Milwaukee, P.O. Box 413, Milwaukee, Wisconsin 53201, USA

This paper shows how state wage surveys can be used to compare the earnings of publicand private sector workers within states. This subject of long-running interest to economists hasgenerated a complex and sometimes contradictory set of empirical results. Surprisingly little,if any, of this academic work has used the publicly available wage surveys that many statesroutinely collect. This paper serves as an introduction to such surveys and an application of theWisconsin State Wage Survey for the purpose of comparing public and private compensation.While the evidence generally indicates public earnings are higher, we are careful to point out thelimitations of our methodology which suggests this conclusion is incomplete.

1. Introduction to the Issue of Comparability

The setting of wage levels in the public sector is unavoidably dominated byadministrative and political processes. At issue is how well these processeswork. Comparability stands as the accepted standard by which economistsmake such judgments. Both equity and efficiency arguments require that publicsector wages be comparable to those of sitnilar workers perfonning similar tasksin the private sector. Equity requires that given workers not be paid differentlyfor identical work simply because their employer is the govemment. Efficiencyrequires that the wage be adequate to attract the needed supply of workers. Toohigh a wage wastes the resources of state and local govemments depriving themof the opportunity to address other objectives or reduce taxes. Too low a wagemakes it impossible for governments to attract workers of the quality neededto provide the services citizetis demand.

* This paper was completed when Franklin was a graduate student working with ProfessorsBelman and Heywood. The authors thank Bill Beutel at the Wisconsin Department of Industry,Labor and Human Relations who provided valuable background information on the Wage Surveyand detailed comments on an early draft. Thanks are also expressed to the UWM workshop Inapplied economics for comments on the early results of this study.

80 D. Belman et al. I Comparing PubUc and Private Earnings

This paper uses the Wage Survey of the state of Wisconsin to compare theeamings of public and private sector workers within that state. The remain-der of this section describes the methodologies used by economists to explorecomparability and discusses the role state wage surveys may play in that ex-ploration. The second section describes the Wisconsin survey, its history andcurrent uses. The third section describes the estimates made and findings onthe issue of comparability. The fourth section discusses several limitations ofthis approach and responds to some of them. The final section draws tentativeconclusions.

Economists have decades of experience comparing the eamings of publicsector and private sector workers. The methodologies can be divided into twobroad classifications. The first examines otherwise similar people through theuse of regression analysis. Using a random sample of workers from each sector,regressions determine an earnings function for each sector. The explanatoryfactors typically include the workers' education, job experience, current jobtenure, race, gender, marital status, and related demographic controls. In eachcase, the average predicted wage for one sector is compared to the actual meanwage in the other sector. Thus, the actual earnings of private sector workersare compared to the wage that was predicted for those workers if they werepaid according to the public sector equation. Similarly, the actual eamings ofpublic sector workers are compared to the wage predicted for those workersif they were paid according to the private sector equation. TTiis process holdsconstant all the determinants of earnings and isolates any remaining differencein wages between sectors.

The process being described is an illustration of the general technique of theOaxaca (1973) decomposition and has been repeated many times in this liter-ature (see Smith, 1976 and 1977; Quinn, 1979; Belman and Heywood, 1988and 1990; Moore and Raisian, 1991). While the evidence from these studiesprohibits simple conclusions, the results suggest that for men state and localeamings are approximately comparable to those for the national private sector.For women, there appears to be a slight advantage for govemment employ-ment. Moreover, that advantage seems to be greater for state than for localgovemment^.

Although the approach of comparing people and standardizing for eamingsdeterminants has been prevalent among academic researchers for the last twentyyears, a second, older methodology, based on the comparison of narrowly de-fined occupations has been used simultaneously. Using this methodology, the

1. The results for females may be the result of less earnings discrimination in the public sectorthan in the private sector.

D. Belman et al. I Comparing Public and Private Earnings 81

wages of public employees such as clerk typists, accountants, or clinical psy-chologists are compared to those of employees in the same occupation in theprivate sector. Among academic researchers, Fogel and Lewin (1974) and,more recently, Moore and Newman (1991) both use this technique to esti-mate comparability. While the regression approach held the nature of peopleconstant, this approach has the advantage of holding the nature of work con-stant. These offsetting advantages have allowed both methodologies to existside-by-side for at least two decades.

A major limitation with many comparisons of specific occupations is thatthey are limited to a narrow set of occupations and do not allow more generalcomparisons across sectors. What would be required to make such a gen-eral comparison is detailed earnings data on a broad set of occupations and anappropriate set of weights with which to aggregate across occupations and con-struct sector wide wage measures. The Wage Survey of the state of Wisconsinmeets both of these needs.

2. The Wisconsin Wage Survey

The Wiscoasin Wage Survey was developed at the beginning of the 198O'swith the first official survey administered in 1984 and released in 1985^. Theprimary impetus for this undertaking was collection of accurate data on prevail-ing wages required for unemployment compensation hearings and to provideemployers with information to guide personnel and compensation practices.Additionally, the survey has been used by the state to provide in state finnsconsidering expansion or out of state firms considering locating in or relocat-ing to Wisconsin with a clear picture of prevailing earnings. The compilationof the survey fell to the Division of Employment and Training Policy of theWisconsin Department of Industry Labor and Human Relations.

Since its inception, the survey has been put to a variety of uses beyond thoseresponsible for its creation. It has been used by the state to investigate the extentof wages near or below the minimum wage. It is commonly used by attorneysand vocational experts in employment and disability suits. It can play the roleof providing an accurate, if partial, picture of the cost of job or employmentloss. It has also been used as evidence in interest arbitration hearings on publicsector earnings. In Wisconsin, local sector workers are subject to final offerarbitration if the two sides cannot reach a voluntary agreement. One of thecriteria listed in the Wisconsin statutes on municipal employee relations to be

2. This start was later than the neighboring state of Iowa but earlier than another nearby state,Illinois. These other state surveys sprung from a common set of needs.

82 D. Belman et al. I Comparing Public and Private Earnings

used in reaching arbitration decisions is the comparability of the offers withprivate sector eamings^. The wage survey has proven useful as it provides dataon both private sector eamings and the eamings of local workers in variousareas of the state. The survey is also used as a base for statistics presentedin several other state publications used for objectives as diverse as long-termindustrial planning and individual job searches.

The survey is conducted every second year with approximately a one yeardelay between the summer mailing of the survey and the public release ofthe data. As of this writing, the most recent Wage Survey available wasconducted in May of 1990 (although the 1992 Survey is due for release inJuly of 1993). More than 7500 private sector employers were surveyed, andfor each occupation the average wage, median wage, typical starting wage,and some information on the dispersion in wages are reported. TTie number ofemployers surveyed and the extent of detail collected has improved with eachnew survey. When the particular occupation is not paid by the hour, an hourlywage equivalent is computed. This is accomplished by determining the basis ofthe salary (weekly, monthly, etc.), detennining how many hours are typicallyworked within that period and dividing the salary by the number of hoursreported. For each occupation, the total number of workers upon which theestimate is based is also provided. This allows employment weighted averagesto be computed across occupations.

In addition to the private employers, the state govemment and most localgovemments (both municipal and school districts) are included in the survey.The consequence is that the private sector survey results are based on a totalof 228,281 workers, the local results are based on 60,030 workers and thestate results are based on 17,703 workers. Thus, approximately between 15and 20 percent of all workers in the state of Wisconsin are included in thepublished survey results'*. The data is presented both for the state as a whole,for 8 Metropolitan Statistical Areas, and for 17 Service Delivery Areas that aredefined in the administration of the Job Partnership Training Act. In addition,the data is presented broken down by broad industrial classification. Thus,for example, the survey allows one to identify prevailing wages for secretariesemployed in the health care industry and working in the area surrounding, andincluding. Green Bay.

The employers surveyed were chosen from the state Unemployment Com-pensation Master File using random sample selection. The survey excludes

3. For more on this and the arbitration system in general see Hill and DeLacenserie (1991).4. These figures were computed by applying the national average of about 40 percent of the

population in the work force to the state population of about 5 million. This yields about 2 million.The sum of surveyed workers in employers in all three sectors is somewhat more than 300,000or 15 percent.

D. Belman et al. I Comparing Public and Private Earnings 83

Table 1

The Occupations which Match Across Sectors

Number of Workersin the Survey

Number of Occupationsin the Survey

Number of Workersin Matched Occupations

Number of MatchedOccupations

Matched Workersas Percent of Total

Matched Occupationsas Percent of Total

Private

228,281

645

97,249

128

42.6

19.8

Local

60,030

168

39,068

86

66.1

51.2

State

17,703

132

14,207

102

80.3

77.3

the smallest firms, usually those with below 10 etnployees, with the specificcriteria varying by industry. This can be contrasted with the U.S. Bureau ofLabor Statistics with conducts the survey of professional, administrative, tech-nical and clerical jobs as a step in the setting of federal compensation. In thatsurvey the minimum establishment size ranges from 50 to 250 depending onthe industry. While sotne indication is given of the range of wages with eachoccupation, this is not fonnalized as a standard deviation which could be usedto generate formal statistical tests.

One start to investigating the comparability of public and private sectorearnings might be to simply calculate the employment weighted average foreach sector using all occupations available in each sector. This proceduregenerates an average private sector wage of $10.57, an average local sectorwage of $9.19 and an average state sector wage of $12.07. There are, however,two important methodological fiaws that make these numbers unreliable: theoccupations on which the averages are based are not the same across sectors,and even those occupations common to both sectors have very different weightswithin the sectors, a topic to be considered in detail in the next section.

In order to make progress comparing the earnings of public and private sectorworkers, the sharply different composition of those sectors must be understood.In earlier work with sharply different data, Belman and Heywood (1988) arguedthat fully a third of public sector workers had no private sector equivalent. Thisset of workers worked in an industry identified as "public administration", andincluded judges, legislators, police, fire fighters and other occupations that didnot have directly similar private sector workers. In addition, there exists a large

84 D. Belman et al. / Comparing Public and Private Earnings

Table 2

Examptes of Matched and Unmatched Occupations

Accountant

AttorneyComputer Programer IIPhysical TherapistNurse AideClertc, MailMaintenance CarpenterCustodianDesigner, ToolSheetmetat WorkerFire InspectorSocial Services Supervisor

Private

X

X

X

X

X

X

X

X

X

X

state

X

X

X

X

X

X

X

Local

X

X

X

X

X

X

X

X

X

The X represents that the occupation has an entry in the Wage Survey for thatsector.

number of occupations in the private sector that are not in the pubiic sectors.For instance, sales and delivery workers and a large number of manufacturingoccupations are simply not part of the typical govemment work force. Thus,no comparison based on positions can, or should, include all workers fromeach sector.

The wage survey provides data separately for the three sectors. It reflects atotal of 645 occupational listings in the private sector, 168 occupational listingsin the local sector and 132 occupational listings in the state sector. Of theseoccupations we identified 128 occupational listings that are in the private sectorand one or both of the public sectors. These 128 occupations include 97,249private sector workers or 42.6 percent of all private sector workers in thesurvey. A total of 86 of these 128 occupations were present in the localsector including 39,668 workers or 66.1 percent of all local sector workers.Finally, 102 of occupations were in the state sector including 14,207 state sectorworkers or 80.3 percent of all state workers. Table 1 provides a summary of theoccupations in which there is overlap'. It is worth isolating a few occupations

5. A very modest amount of judgment was required to identify matching occupations. Of the128 we identify as matched, 125 are exact matches requiring no judgment. Two additional occu-pations were included as they seemed to be in each sector but at different levels of aggregation.For both registered nurses and laborers, the private sector included different classifications forspecialties while the public sectors presented a single summary measure. In order not to lose theseoccupations we generated a summary earnings measure for the private sector as an employmentweighted average of various specialties. The final occupation included resulted from an effort toinclude the highest ranking executive officers. This category has very few workers and probably

D. Belman et al. I Comparing Public and Private Earnings 85

that overlap and that don't just to set the stage. These are presented in Table 2.This evidence suggests that the Wage Survey will provide an important,

if partial, picture of the relative earnings. By limiting our attention to thematched occupations we will be able to compare the earnings of over half thelocal sector labor force and nearly two-thirds of the state sector labor force.We will be unable to make comparisons for the remaining portions of localand state workers.

3. The Basic Research Design

The fundamental estimates of this section were generated by a series of ag-gregations across the separate sectors and then by comparisons of the sectoralaverages. The comparisons will use the mean and starting wages from the sur-vey. We recognize the potential importance of the median wage but in all ofour investigations we find the median to be uniformly close to the mean, sug-gesting little or no skewness^. A full data appendix upon which all calculationsin this paper are based is available from the authors upon request.

As a first estimate, the earnings of state workers are compared to those ofprivate sector workers. In 19 occupations the average earnings of private work-ers are higher while in 83 occupations the average earnings of state workers arehigher. Weighting each of the 102 occupatiot\s by the number of employees inthe separate sectors (state mean wages weighted by state employments acrossthe occupations and private wages weighted by private employments) yields anaverage state wage $1.58 above that of the private sector wage and a percent-age differential of over 15 percent as shown in Table 3. Similar comparison oflocal and private sector workers indicates that the average mean local sectorwage is $1.16 below the private wage, a percentage differential of negative11.2 percent. Again, this is presented in the top of Table 2. In each case,the weights were generated from the responses to the wage survey and theiraccuracy depends on the success of the survey's selection methodology^.

It should be immediately recognized that these estimates are suspect as ac-curate comparisons of what workers from one sector would earn if moved tothe other. This follows as each of the weighted averages is a function not

has little influence on the outcome. The category of president was combined with that of mayoror executive officer.

6. Indeed, use of the median never changed the relative rankings of the estimated wages acrosssectors and usually changed percentage differences only 2 or 3 percent.

7. While detailed occupational response rate studies have not been done, the distribution acrossthe occupations is, indeed, similar to that of the state as indicated in the more comprehensive"Wisconsin Projections 1988-2000".

Employment in theseparate sectors

Employment in theprivate sector

Employment in thestate sector

Employment in thelocal sector

10.49

10.35

10.49

10.35

10.25

7.88

1.58

-1.16

1.67

0,30

1,82

15,1

-11,2

15,9

2,9

17,8

86 D. Belman et al. I Comparing Public and Private Earnings

Table 3

State Weighted Averages of the Occupational Mean Wages

Weighting Private State Local Differential Percent

12,07

9,19

12,16

10,65

12,07

9,19 1,31 16,6

The percentage differential is calculated as the differential as a percent of the private wagefigure,

only of the underlying occupatiotial meatis but also the value of the weights ineach sector. In other words, one reason the state wage may be so much higheris that the state employs a larger proportion of high wage professionals. Thiscould result in a large estimated differential even if the underlying occupationalmeans were identical between the two sectors.

Several strategies exist to avoid this problem. One would be to calculate thedifferential at the occupational level and then aggregate. While this provides apartial answer, the proper aggregation weights remain in doubt. An alternativeapproach is to generate the aggregate means for each sector using identicalweights. This approach holds constant the weights across sectors and suggeststwo alternatives schemes. The first uses the private sector weights while thesecond uses the public sector weights. These two weighting schemes haveimmediate interpretations. Usitig the private sector weights indicates what thatwork force would earn if they were paid the occupational means of the publicsector. Similarly, the public sector weights indicate what either the state orlocal work forces would earn if they were paid according to the private sectoroccupational means.

Table 3 presents each of these alternative weighting schemes for both theprivate-state and private-local comparison. As before, the weights are from theoccupational employment totals in the survey. The state comparisons show onlymodest change. That which uses the private sector weights indicates a statewage premium of 15.9 percent while that using the state sector weights indicatesa state wage premium of 17.8 percent. These figures are each somewhat higherthan the previously estitiiated figure of 15.1 percent. As there is no particular

D. Belman et al. I Comparing Public and Private Earnings 87

reason to prefer one scheme to the other, a simple average of 16.9 percentindicates roughly how much damage is done by using the separate weights ofthe different sectors.

The comparison between the local and private sectors changes more dramat-ically once the weights used no longer vary with the sector. While the initialdifferential suggested a substantial private sector premium, that result vanishesonce the weights are held constant. Using the private sector weights, the pre-mium reverses and indicates that if the private sector work force received theoccupational means of the local sector they would enjoy eamings of approx-imately 2.9 percent higher. Even more dramatically, the local sector weightsindicate that the premium is 16.6 percent in favor of the local workers. Thisrepresents the extent to which the earnings of the local work force would fallif they were paid the occupational means of the private sector.

The reversal of the sign in the private-local sector comparison suggests thatthere exists a variant on the famous statistical phenomenon known as Simpson'sparadox. In the extreme version of this paradox, it would be possible for eachand every occupation to have a larger wage in one sector but for the other sectorto have a higher average wage because of the difference in weights acrossthe two sectors'^ To check whether the weights are sufficiently asymmetricto generate this extreme version of the paradox, the local-private comparisonwas divided into two separate comparisons. In the first, all occupations wereexamined in which the local wage was greater than the private wage. Theaverage wage was computed for each sector using the own sector weights andthe percentage differential is shown in the first row of Table 4. The positivedifferential shows no reversal of the sort implied by the paradox. The procedureis repeated in the second comparison for all occupations in which the privatewage is greater than the local wage. The negative differential again shows noreversal and no evidence of the paradox.

Table 4 also shows the differentials calculated separately for each set ofoccupations using the weights of a single sector as a base. Again, no reversalis indicated. The consequence of this investigation is that the original negativelocal differential in Table 2 results from the different weights given by eachsector to the occupations in which the private wage was higher and in whichthe local wage was higher. The asymmetry of weights was not severe enoughto generate the extreme form of the paradox. Also interesting, is that Table 3makes clear that the private sector pays higher wages in those occupations that

8. Essentially what is required for this to happen is for the occupations with lower earningsto have large weights in one sector and those with high earnings to have large weights in theother sector. For more on the conditions necessary to observe this extreme version of Simpson'sParadox, see Thronton and Innes (1985) and Samuels (1993).

88 D. Belman et al. I Comparing Public and Private Earnings

Table 4

Testing for Simpson's Paradox

Local Mean Wages Greater than Private Mean Wages

Unweighted

Weighted by Private

Weighted by Local

Private Mean

Unweighted

Weighted by Private

Weighted by Local

Private

9.71

8.32

7.08

Local

11.50

9.87

8.83

Difference

-1.77

-1.55

-1.75

Wages Greater than Local Mean Wages

Private

17.77

18.08

16.01

Local

14.05

13.58

12.85

Difference

3.72

4.50

3.16

Percent

-18.2

-18.6

-24.7

Percent

20.9

24.9

19.7

Table 5

Statewide Weighted Averages of the Occupational Starting Wages

Weighting Private State Local Differential Percent

Employment inseparate sectors

Employment inprivate sector

Employment instate sector

Employment inlocal sector

the

the

the

the

8.45

8.28

8.45

8.28

8.32

6.60

9.18

9.25

9.18

8.01

9.49

0.73

-0.27

0.80

1.21

0.86

8.6

-3.3

9.5

14.6

10.3

8.01 1.41 21.4

The percentage differential is calculated as the differential as a percent of the private wagefigure.

generally pay more and the local sector pays more in those occupations thatgenerally pay less.

Table 5 duplicates the presentation of Table 2 using the occupational startingwages. As expected, the starting wages are uniformly lower than the meanwage. The private sector starting wage is about 80 percent of its mean, thestate sector starting wage is 76 percent of its mean, and the local startingwage is 87 percent of its mean. TTie comparison of starting wages whichuses the separate weights from each sector are shown at the top of the tableand indicates that state starting wages are 8.6 percent higher than those in the

Employment m theseparate sectors

Employment io theprivate sector

Employment in thestate sector

Employment in thelocal sector

11.79

11.60

11.79

11.60

10.82

8,87

0.28

-0.99

0,88

-0.01

1.25

2,4

-8.5

7,5

0,0

11,6

D. Belman et al. I Comparing Public and Private Earnings 89

Table 6

Milwaukee Weighted Averages of the Occupational Mean Wages

Weighting Private State Local Differential Percent

12.07

10,61

12.67

11,61

12.07

10,61 1,74 19,6

The percentage differential is calculated as the differential as a percent of the private wagefigure,

private sector while those in the local sector are 3.3 percent below those in theprivate sector.

Again, these numbers are misleading as the differences could refiect differ-ent distributioas of employment across the occupatiotis rather than differentearnings within the occupations. The next section of Table 4 presents thedifferentials using the private sector weights. The state advantage over the pri-vate sector appears to be 9.5 while that in the local sector reverses sign again.The small negative local differential becomes a large positive differential of14.6 percent.

The public sector weights are presented in the final rows of Table 5. Thestate premium using the state weights is 10.3 percent. A simple average of thiswith the private sector differential suggests a premium of 9.9 percent. On theother hand, the local sector weights suggest that the local sector premium overthe private sector is 21.4 percent. The two local sector estimates (with local andprivate weights) average to positive 18.0 percent, a far cry from the negativedifferential of 3.3 percent. This difference again highlights the importance ofusing the same weights in order to hold the composition of the very differentsamples constant.

Thus, the comparison of public and private occupational wages both at thestarting level and for the means suggest a consistent story with public earningsabove those in the private sector. The average state differential (holding theweights constant) is 16.9 percent for the mean wage and 9.9 percent for thestarting wage. The average local differential (holding the weights constant) is9.8 percent for the mean wage and 18.0 percent for the starting wage.

90 D. Belman et al. I Comparing Public and Private Earnings

It remains difficult to properly isolate the source of the differences withinthe general pattern. The fact that the state advantage is smaller at the startinglevel might indicate a steeper experience profile for state workers. Thus, anadditional year of tenure could be better rewarded in the state sector. Alterna-tively, state employees may stay in their position and with their employer forlonger periods of time. Thus, the state mean looks further above the privatesector mean than do the starting wages because the average state worker hasmore years of tenure. The same two arguments remain for the local sectordifferentials but they are reversed. The fact that the mean differential is lowerthan the starting differential could reflect either a flatter earnings profile or lessaverage tenure.

4. Additional Considerations

While the Wage Survey provides a valuable comparison of the extent ofcomparability, there are a number of caveats that argue for circumspection.We will respond to one of the major caveats, however, others are beyond thescope of investigation that can be carried out with the Wage Survey.

First, a fifth of the state workers and a third of the local worker are excludedfrom the comparison. Making wage comparisons for these excluded workersis made difficult by the fact that they are in occupations not present in theprivate sector. Attempting to establish comparability in these occupations mustbe based either on a more detailed comparison of the particular tasks that thejobs require, or on the basis of whether or not the earnings are similar to whatsimilarly educated and experienced people earn in the private sector.

Second, there could exist substantial variation within the duties actually per-formed within a given occupation despite the seeming narrowness of the oc-cupational categories. Indeed, the survey anticipates this problem to someextent by providing precise descriptions rather than occupational codes alone.Nonetheless, there could exist substantial variation in the quality of the worksetting even if the duties are functionally similar. To take an extreme, a securityguard in a department store and a security guard in a state prison might sharean occupational code but few would think their wages should be comparable.Determining the extent of the bias introduced by such aggregation must waitfor even more disaggregate wage surveys which present even finer classifica-tions of the occupations in each sector. Absent such classifications we are leftto speculate as other arguments might present the opposite bias. State and localjobs may provide more job security, or in the case of professionals (lawyers orclinical psychologists) the public earnings do not include a risk premium forbeing an independent business person.

D. Belman et al. I Comparing Public and Private Earnings 91

Third, the results presented from the Wage Survey are difficult to reconcilewith the typical regressions surveys that attempt to standardize for humancapital and other demographic characteristics. In general, most nationwidestudies indicate that when combining women and men in the same samplestate workers enjoy a premium of at most a couple of percentage points. Localworkers are often shown to have a substantial negative differential. Indeed,Belman and Heywood (1994) have taken a representative sample of Wisconsinworkers for the years 1989, 1990 and 1991 (centered on the year of the survey)and have found the state premium to be a modest 2.34 percent and the localdifferential to be —6.36 percent. That study, and others like it, standardizefor years of education, experience, race, gender, the broad occupation andother demographic variables. Unlike studies based on detailed occupations,such as this one, that study presumes that similar workers, those with similarcharacteristics, should be paid similarly.

The stark difference in results between methodologies suggests several pos-sible reconciliations. The education of public workers may be greater than thatof private sector workers within the same occupational categories. Even if thisconjecture were true, evaluating the consequences in terms of comparabilitydoes not immediately follow. If the extra education is needed to perfonn therequired tasks in the manner demanded by citizens, few would object. If theextra education is a consequence of the fact that higher public earnings attractan unnecessarily educated work force, objections may be warranted. Indeed,in the latter circumstance comparability would dictate not allowing education"to explain" the difference in earnings between the public and private sector.While it is known that the public work force in Wisconsin has more educationon average (see Belman and Heywood, 1994), it is not known whether this isa function solely of the distribution of workers across occupations or if it alsoremains true within occupations.

A similar issue, requiring even more judgment, follows from the racial andgender differences between the work forces of the separate sectors. Again,the occupational comparisons completed in Section 3 made no attempt to stan-dardize for these differences. The regression earnings approach allows thesevariables to explain a portion of the difference in earnings between sectors.Belman and Heywood (1994) have shown that the state sector has approx-imately twice as much minority representation as the private sector and thelocal sector may have more than that. If minorities are discriminated against,perhaps it should not be reproduced in the public sectors. Thus, if all publicsector workers received the same earnings as white males in the private sector,occupational earnings differences could easily result. Thus, the public policy ofgovernment wage determination might appropriately deviate from the standardof strict comparability in this regard.

92 D. Belman et al. I Comparing Public and Private Earnings

Fourth, no effort has been made in our earlier comparisons or, for that mat-ter, in the typical earnings regression equations, to account for differences innonwage compensation. These fringe benefits might include pensions, healthand life insurance benefits provided by employers. TTiese may differ both inthe odds of being provided across the sectors and in the generosity of the ben-efits where provided. A sophisticated analysis that incorporates these variouscomponents might result in compensation premiums that differ substantiallyfrom those we have presented for earnings alone.

Fifth, the results we present are generated from a single cross-section inthe year 1990. The dynamics of private and public sector earnings may varyover the business cycle. It is possible that the heavily unionized public sectorshows less cyclical variation in earnings than does the private sector. This couldgenerate apparent earnings differences that would disappear over the long run^.

Sixth, all state workers and many local workers are employed by large or-ganizations. It is well known that larger private sector employers pay morethan smaller employers. To the degree that the public sector requires similarcompetence and competes for employees with large private finns, the wagesof these larger firms should be the appropriate benchmark. Belman and Hey-wood (1990) have shown that holding the size of the firm constant in typicalearnings regressions results in smaller government differentials. Indeed, thelarge federal differential falls to insignificantly different from zero when con-ditioned on firm and establishment size. The fact that the Wisconsin WageSurvey does not control for firm and establishment size might generate the 15to 19 percent advantage we report. The smaller private sector firms may begenerating much of this advantage but these finns should perhaps be excludedas they are not comparable with large governmental employers. Although theinformation on private sector firm size is available in the state wage surveyquestionnaires, it is not included in the tabulations^".

Finally, the argument is often made that one reason state and local earningslook so large is that workers in those sectors are concentrated in the state capitaland the major metropolitan areas. In the case of Wisconsin, the argumentwould contend that public sector workers are disproportionately concentratedin the urban areas of Madison and Milwaukee. The average earnings and theaverage cost of living are higher in these communities than in the smallercities and towns of the remainder of the state. Thus, the proper examination

9. The Wisconsin Department of Industry, Labor and Human Relations has begun its ownpublic-private wage comparison associated with each wave of the survey, tf this continues to bedone, a time-series of estimates will evolve,

10, Internal comparisons done by the Department of Industry, Labor and Human Relationssuggest that the positive wage advantage is smaller when compared to large firms (250 workersor more) but that it does not vanish. See "A Comparison", Technical Report, July 1991,

D. Belman et al. I Comparing Public and Private Earnings 93

of comparability would be to focus on the occupational earnings of public andprivate sector workers in the communities in which they actually work. A statewide comparison biases the earnings differentials such that what is really anappropriate cost of living difference appears as a premium for state and localworkers.

Unlike the early caveats, in which we could do little beyond make the point,here we can explore the significance of the argument by using another featureof the Wage Survey. One of the areas for which the wage figures are brokenout separately is the Milwaukee Statistical Area. This allows us to apply themethodology used for the mean wage comparison for the entire state to thesingle area of Milwaukee.

There were 114 private sector occupations that matched with either state orlocal occupations within the Milwaukee area. This included the same numberof state occupations as before as they are not specific to locale and included 65matched local sector occupations which were specific to the Milwaukee MSA.Table 5 presents the mean wage comparisons in the fashion analogous to thosefor the state-wide sample. The first two rows demonstrate that when using theactual weights of each sector the state and private wage seem to be similar(state workers state-wide seem to be paid like their private sector Milwaukeecounterparts) and that local government workers in Milwaukee eani substan-tially less than their private sector Milwaukee counterparts.

Altering the comparisons to hold the weights constant with those of theprivate sector indicates that state workers receive a premium of 7.5 percent andthe earnings of local workers are virtually identical to those of private sectorworkers. Using the government sector weights indicates a state premium of11.6 and a local premium of 19.6 percent. Averaging the two bases gives anaverage 9.6 percent state premium and an average 9.8 percent local premium.The state numbers are, indeed, less extreme than those presented with the state-wide sample but they are qualitatively similar. The local sector premium isremarkably similar to that calculated on the state-wide sample.

5. Conclusions

A wide variety of states have found that there is substantial demand forconducting state wage surveys. These surveys have found many uses by bothgovernment and private business people. Surprising little use has been madeof the surveys for asking traditional academic questions. The extent of wagecomparability between the public and private sectors is an academic questionand a practical policy issue for which the surveys can be used.

94 D. Belman et al. I Comparing Public and Private Earnings

In a series of aggregations across the matched set of occupations in Wisconsinwe found consistent evidence of a premium for the state and local govemmentworkers when compared to their private sector counterparts. This did not appearto be a function of the concentration of govemment workers in high-wage urbanareas. We also suggested a number of shortcomings with our methodology thatindicated reasons for caution. Yet, the exercise did indicate the type of projectwhich can be completed with the detailed infonnation collected in such wagesurveys.

References

Belman, Dale and John Heywood (1994). State and Local Government Wage Differentials: AnIntrastate Analysis. Journal of Labor Research, forthcoming.

Belman, Dale and John Heywood (1990). The Effect of Establishment and Firm Size on PublicWage Differentials. PubUc Finance Quarterly, 18, 221-235.

Belman, Dale and John Heywood (1988). Public Wage Differentials and the Public Administration"Industry". Industrial Relations, 27, 386-394.

Fogel, Walter and David Lewin (1974). Wage Determination in the Public Sector. Industrial andLabor Relations Review, 27, 410-431.

Hill, Marvin Jr. and Emily DeLacenserie (1991). Interest Criteria in Fact-Findingand Arbitration:Evidentiary and Substantive Consideration. MarqueUe Law Review, 74, 399^49.

Moore, William and Robert Newman (1991). Government Wage Differentials in a MunicipalLxibor Market: The Case of Houston Metropolitan Transit Workers. Industrial and LaborRelations Review, 45, 145-153.

Moore, William and John Raisian (1991). Government Wage Differentials Revisited. Journal ofLabor Research, 12, 13-34.

Oaxaca, Ronald (1973). Male-Female Wage Differentials in Urban Labor Markets. InternationalEconomic Review, 14, 693-709.

Quinn, Joseph (1979). Wage Differentials Among Older Workers in the Public and PrivateSectors. Journal of Human Resources, 14, 42—62.

Samuels, Myra (1993). Simpson's Paradox and Related Phenomena. Journal of the AmericanStatistical Association, 88, 81-88.

Smith, Sharon (1977). Government Wage Differentials. Journa/o/t/rfcan £co«om/cx, 4, 179-197.Smith, Sharon (1976). Government Wage Differentials by Sex. Journal of Human Resources,

11, 185-199.Thornton, Robert and Jon Innes (1985). On Simpson's Paradox in Economic Statistics. Oxford

Bulletin of Economics and Statistics, 47, 387-395."1990 Wage Survey - State of Wisconsin" (1991). Department of Industry, Labor and Human

Relations, State of Wisconsin."1990 Wage Survey - Milwaukee MSA" (1991). Department of Industry, Labor and Human

Relations, State of Wisconsin."1990 Wisconsin Wage Survey Public vs. Private Sector Wages: A Comparison". A Technical

Report, (July 1991). Department of Industry Labor and Human Relations, State of Wisconsin.